howardpen9/kimi-code-mcp
View on GitHub ↗MCP server for Claude Code × Kimi K2.5 (256K context) — delegate bulk codebase analysis to Kimi, save 90% on token costs. Session caching, parallel agents, TypeScript.
60 ★10 forksTypeScriptUpdated 1mo ago
What you need to know
An MCP server that delegates bulk codebase reading from Claude to Kimi Code (256K context) to cut Claude-side token cost by roughly 90%, with API and CLI backends.
Install
npm install -g kimi-mcp-server Optional: install the Kimi CLI with curl -L code.kimi.com/install.sh | bash and run kimi login
Usage
- •Ask Claude to analyze a large codebase with kimi_analyze
- •Use kimi_verify to cross-check code or diffs as an independent reviewer
- •Resume sessions with kimi_resume
- •Check backend status with kimi_status
Key features
- ✓8 tools: kimi_analyze, kimi_query, kimi_verify, kimi_list_sessions, kimi_resume, kimi_status, kimi_cache_status, kimi_cache_invalidate
- ✓Two backends: direct API for query/verify, local CLI for codebase reading
- ✓Output control with detail_level, max_output_tokens, and include_thinking
- ✓Session caching with warmup and reuse
- ✓Token compression ratios from 5x to 100x depending on detail level
Best for
Claude Code users with large codebases who want to cut token costs while keeping Claude in charge of decisions and edits.
Caveats
- ⚠Requires an active Kimi Code subscription
- ⚠Kimi Code API and Moonshot API keys are not interchangeable
- ⚠Skip Kimi for small codebases under ~10 files
Clients: Claude Code · Any MCP-compatible client
Reviewed 2026-08-11
Topics
ai-agentai-coding-assistantclaude-codecode-analysisdeveloper-toolskimikimi-codekimi-k2llmmcpmcp-servermodel-context-protocolmoonshot-aitypescript
- Stars
- 60★
- Forks
- 10
- Language
- TypeScript
- License
- MIT
- Created
- 2026-03-02
- Last push
- 2026-07-28